Reproduce “Preference-Guided Adaptation for Open-Vocabulary Semantic Segmentation via Prompt Disagreement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hyun-Kurl Jang, Jihun Kim, Kuk-Jin Yoon
The method adapts open-vocabulary segmentation to specialized imagery by generating competing masks from different prompt wordings, asking which better matches it in a localized region, and updating with preference and consistency losses. On MESS, CAT-Seg-L improves from 35.26 to 45.88 mIoU overall, below matched dense-mask supervision at 46.67 except on Engineering. The benchmark oracle used ground truth; human adaptation was not tested.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.